Received-SPF: pass (google.com: domain of linux-kernel+bounces-76956-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII;
 format=flowed
Content-Transfer-Encoding: 7bit
Date: Thu, 22 Feb 2024 17:07:23 +0000
From: ns@tfwno.gf
To: Masahiro Yamada <masahiroy@kernel.org>, Nathan Chancellor
 <nathan@kernel.org>, Nicolas Schier <nicolas@fjasle.eu>,
 linux-kbuild@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: make install doesn't install proper files for some architectures
User-Agent: Roundcube Webmail/1.4.15
Message-ID: <06e09da69e6c34d03a303c11088d9ffd@tfwno.gf>

Greetings,

The kernel's build system has an install target. Most of the time this
works fine, but it is limited to only one file, which can make it busted
for some arches.

One example of a target where it is busted is a generic ppc32 kernel,
where there are plenty of supported platforms, some of which can boot
perfectly fine through Open Firmware & some which are from bad, bad
times o' yore where you had to embed the DT straight into the kernel. &
some more on top of that. When you run make install for ppc32, it only
installs the uncompressed vmlinux. This is almost certainly not the
correct decision for _any_ of the platforms in question, & it totally
leaves out the properly wrapped images that many of these platforms
absolutely need in order to boot (e.g. PS3 & Wii, which need an image
with an embedded DT).

This is not really great. By contrast, the all target built all the
specific images that my config selected. I didn't have to read the
arch's boot Makefile to figure out what files I have to build. Is there
any target which does the same but for installing every image the all
target produced? It'd be a much better experience than leaving the
details of every arch & bizzare platform supported by it up to (most
likely incomplete) scripts, when the Makefiles already have all that
information properly written. I suppose it'd also come in handy if e.g.
x86 had a saner image format [1] in the future or something of the sort.

[1]: 
https://source.denx.de/u-boot/u-boot/-/blob/master/doc/usage/fit/x86-fit-boot.rst